Introducing smolagents: simple agents that write actions in code.
Hugging Face Blog
Hugging Face released smolagents, a library for building AI agents that execute actions by writing code rather than JSON, with support for any language model including open-source options. The library contains approximately thousands of lines of code and supports execution in sandboxed environments via E2B, while enabling tool sharing through the Hugging Face Hub. This approach allows language models to handle complex, real-world tasks with unpredictable workflows that would be difficult to manage with predetermined control flows.